Bayesian Off-Grid Near-Field Channel Estimation

We estimate the uplink channel of an extra large-scale multiple-input-multiple-output (XL-MIMO) system with hybrid architecture at the base station. Due to large Rayleigh distance in XL-MIMO systems, the channel exhibits near-field (NF) properties, and becomes burst sparse in the polar domain. We design a dictionary learning-based gridless NF Bayesian learning (GN-BL) algorithm to exploit this burst sparsity. Along with the polar domain channel, we also estimate its angles and distances, which improves the estimation accuracy. We show that our GN-BL algorithm has a lower normalized mean squared error than the state-of-the-art NF channel estimation algorithms. The GN-BL algorithm, due to its architecture, also has lower time complexity than other sparse Bayesian learning techniques.